Cheap CPI Markets: When Cheap Installs Cost You More

Because we could not find a single country-level CPI figure that traces back to a source we can open, read and stand behind — every candidate led to an aggregator article quoting another aggregator article. So we are not printing one, and the reason is worth more to you than the table would have been.

The pattern is easy to see once you look for it. A number appears in a listicle. A second listicle cites the first. A third cites the second and rounds it. Within a year the figure is everywhere and its origin is nowhere. Nobody publishes the auction it came from, the date range, the ad network, the platform split, the campaign objective, or whether the installs were counted by the store or by an attribution provider. Without those, a CPI number is not a benchmark; it is a rumour with a currency symbol on it.

There is a deeper problem than sourcing. CPI is not a property of a country. It is the output of an auction you are competing in, against advertisers bidding for the same attention with different products, different budgets and different tolerance for loss. Your CPI in a given market is set by your creative, your category, your store listing conversion rate, your bid strategy, your seasonality and your competitors' balance sheets. Two apps running in the same country in the same week routinely pay prices that are not in the same range.

So the question "which countries have the cheapest CPI" has no stable answer, and a table pretending otherwise would mislead you in the most expensive way possible: it would make a budget decision feel researched. The answerable question is different. Given what a market pays back, what install price could you afford there — and can you actually measure whether you hit it? That is what the rest of this article is about, and it is also how we approach market selection in our user acquisition work.

An install is an event, not an outcome, and a cheap one buys you the same event at a lower price with no promise about anything downstream. The trap in low-CPI expansion is that the cost side of the equation improves immediately and visibly, while the value side degrades slowly and invisibly.

Think about what has to happen after the install for the spend to have been worth it. The user has to open the app. They have to complete onboarding in a language they read comfortably. They have to find the value proposition credible in their context. They have to return. Then, depending on your model, they have to see enough ad inventory at a price advertisers will pay in that market, or they have to reach a paywall priced in their currency and decide it is worth it.

Each of those steps is a market-specific multiplier, and each one can be lower in exactly the places where installs are cheapest. That is not a coincidence — it is the same underlying condition expressed twice. Advertiser demand is low in a market because the expected value of a user there is low, which is why the auction clears cheaply. The cheap price is a signal about the value, not an escape from it.

Across the apps we have managed since 2013, the geo expansions that failed rarely failed on CPI. They failed on the assumption that a user acquired in a new market behaves like a user acquired in the home market. Retention curves, session depth, payment method availability and paywall conversion all shift, and they shift independently of each other. Our LTV and CAC calculator guide walks through the arithmetic that makes this concrete.

Because the stores themselves adjust your price downward in those markets, so the same purchase produces less revenue there than it does at home — by design. This is documented, and it is the single most overlooked mechanic in geo expansion planning.

On the App Store, Apple's guidance on setting a price explains that you set a price in a country or region you are familiar with "as the basis for automatically generating prices across the other 174 storefronts and 43 currencies", and that automatically generated prices "account for foreign exchange rates and certain taxes, and follow the most common pricing convention for each country or region". Apple also states that it periodically updates prices in certain regions based on changes in taxes and foreign exchange rates, and that it "will never change the price in your base country or region" while notifying you in advance of changes on other storefronts.

Google Play works on the same principle. Play Console's pricing documentation states that Play converts your price to the local currency, adds tax in select countries, and applies locally relevant pricing patterns and valid exchange rates for the date on which you set the price. It also states that where a local currency is not supported, the price you enter is used and a price in USD or EUR is generated, and that the currency used in each country is set and cannot be changed.

Read that back with a UA budget in your hand. If you sell a subscription and let the stores generate local prices, your gross revenue per paying user in a lower-income market is lower before a single behavioural difference enters the picture. Then behaviour compounds it: conversion to paid, payment method success and renewal rates all vary by market.

The same argument holds for ad monetisation, in a different form. Your effective revenue per impression is set by what advertisers will bid for attention in that geography — and if install prices there are low because advertiser demand is low, your ad revenue per user is exposed to the same weakness. Cheap traffic and cheap monetisation are two readings of one instrument. If you are deliberately setting prices by market rather than accepting generated ones, our note on purchasing-power-based price localisation covers how to do it without destroying the base price.

Not necessarily, and the two platforms define "country" differently enough that your ads data and your store data can disagree without either being wrong. This is the first thing to check when a new market's numbers look strange.

Google Play is explicit. Its documentation on distributing app releases to specific countries states that "country targeting is based on the user's Play country (that is, where their account is registered), not their current location". A user physically in one country whose Play account is registered in another is, to Play, in the registration country. Availability and release targeting follow the account, not the handset.

Google Ads defines it differently again. Its guidance on targeting ads to geographic locations states that location targeting "is based on a variety of signals, including users' settings, devices, and behavior on our platform", describes the result as Google's best effort to serve ads to users who meet your location settings, and says plainly that "100% accuracy is not guaranteed in every situation". The broader setting reaches people who are in your targeted locations or have shown an interest in them; the presence setting is the one to consider when you only want users actually in those locations.

The practical consequence is that a cheap market can be cheap partly because you are reaching an audience adjacent to the one you intended, on interest signals rather than presence. That audience is not fraudulent and the installs are real. They are simply not the cohort your business case was written about, which is why the retention curve does not match the plan.

Often it cannot, because both the bidding systems and the modelling tools you would use to evaluate a market have documented volume thresholds that a small test in a new country does not clear. This is the step most teams skip, and it is the reason a lot of geo tests produce a confident conclusion from data that could not support one.

Take value-based bidding first. Google's documentation on Target ROAS bidding lists minimum requirements that differ by campaign type, and for App campaigns it states the requirement as at least 10 conversions every day, or 300 conversions in 30 days. If your new-market test cannot produce that conversion volume, you are not choosing between install-optimised and value-optimised bidding — you only have one option, and it is the one that optimises for the event you already know is a poor proxy for value.

Now take prediction. Google Analytics 4's predictive metrics — purchase probability, churn probability, predicted revenue and in-app purchase probability — require model training, and the predictive metrics documentation states the prerequisite as at least 1,000 returning users having triggered the relevant predictive condition and at least 1,000 returning users not having triggered it, over a seven-day period within the last 28 days, with model quality sustained over time. GA4 defines predicted revenue as the revenue expected from all purchase key events in the next 28 days from a user active in the last 28 days.

Those thresholds are not obstacles to work around. They are a statement about how much signal a market has to generate before any model can say something reliable about it. A two-week test in a low-volume country produces a number, and the number will be precise, and it will not be trustworthy. If you need an early read on a market before the volume exists, a structured soft launch is the honest instrument — our soft launch strategy guide covers how to design one that answers a question rather than decorating a decision.

It raises the pressure on your verification, because low payout markets are where the economics of manufacturing an install work best for a bad actor. The exposure is not evenly distributed across channels, and it is not identical to what the ads platform already filters for you.

Google's documentation on invalid traffic defines it as clicks and impressions on ads that are not the result of genuine user interest, including intentionally fraudulent traffic and accidental or duplicate clicks. It states that invalid traffic is detected by Google's monitoring systems, that advertisers are not charged for invalid clicks or impressions, and that the invalid clicks column shows traffic the automated systems have already detected and filtered in real time.

That protection is meaningful but bounded. It concerns ad interactions on Google's own inventory. It says nothing about the quality of installs sourced from an incentivised network, an offerwall, or a reseller two layers below the partner you signed with — which is precisely where the cheapest inventory in the cheapest markets tends to live.

The defence is behavioural, not forensic. Fraudulent installs are cheap to manufacture and expensive to make behave: they arrive, they are attributed, and then they do nothing. Segment every new market by source down to publisher level and compare day-1 and day-7 retention, session depth and any post-install event that is difficult to fake. A source whose installs never reach a second session is telling you what it is. Our guide to mobile ad fraud prevention goes through the signatures in detail.

Start from your own organic users in that market, because they are the only unbiased sample of how your product performs there and they cost you nothing to observe. Almost every app that is considering a market already has some presence in it, and that presence is more informative than any external benchmark.

The output of this process is not a CPI figure. It is a maximum affordable install price for that market, derived from your revenue per user and your payback tolerance. That number is defensible because every input is yours. It also changes the negotiation entirely: instead of asking a network what installs cost in a country, you tell them what you can pay and let them say whether they can deliver at that price.

Compare that with our teardown of what an app install actually costs in India, which works through the same logic in a single market rather than in the abstract.

Whether the market returns your money faster than you need to spend it again — because a market with a good lifetime return and a slow payback can still bankrupt the campaign that funds it. Cheap installs make this worse, not better, and that is the counterintuitive part.

A low-CPI market with a low revenue per user typically has a longer payback period than a high-CPI market with a high revenue per user, even when both eventually reach the same ratio of value to cost. Cheapness buys volume, and volume in a slow-paying market ties up more cash for longer. You can be profitable on a spreadsheet and out of working capital at the same time.

The comparison that matters is not between two countries. It is between one country's payback period and your own cash cycle.

This is why we treat market selection as a finance question wearing a marketing costume. The full mechanics, including how to work out how long you can carry a cohort, are in our piece on UA budgets and cash payback.

Write down the affordable install price for the market before you look at any quoted price, so the quote cannot anchor your judgement. Everything else in this article exists to make that one number defensible.

If the affordable price you derive is below what any legitimate source will deliver for, the market is not cheap. It is unaffordable at your current unit economics, which is a product and pricing problem rather than a media buying one. That is a far more useful conclusion than a table would have given you, and it points at work you can actually do.

If you want a second opinion on a market before you commit budget to it, send us the organic numbers you already have and we will tell you what they imply about the affordable install price. It is usually a short conversation, and it is a great deal cheaper than finding out in the auction.